API Pack: A Massive Multi-programming Language Dataset For API Call Generation

Guo Zhen, Soria Adriana Meza, Sun Wei, Shen Yikang, Panda Rameswar. Arxiv 2024

[Paper] [Code]    
Fine Tuning GPT Has Code Model Architecture Pretraining Methods RAG Tools Training Techniques

We introduce API Pack, a massive multi-programming language dataset containing more than 1 million instruction-API call pairs to improve the API call generation capabilities of large language models. By fine-tuning CodeLlama-13B on 20,000 Python instances from API Pack, we enable it to outperform GPT-3.5 and GPT-4 in generating unseen API calls. Fine-tuning on API Pack also facilitates cross-programming language generalization by leveraging a large amount of data in one language and small amounts of data from other languages. Scaling the training data to 1 million instances further improves the model’s ability to generalize to new APIs not used in training. To facilitate further research, we open-source the API Pack dataset, trained model, and associated source code at https://github.com/zguo0525/API-Pack.

Similar Work